Deep reinforcement learning-based low-speed vehicle following decision-making method

Abstract

The invention discloses a deep reinforcement learning-based low-speed vehicle following decision-making method, which is implemented in the following manner: at first, receiving position, speed and acceleration information of front and back vehicles as an environmental state in real time through the Internet of vehicles, and expressing a present state and behavior of an unmanned vehicle; then, constructing an Actor-Critic framework-based deep reinforcement learning structure; and finally, selecting, by Actor, an appropriate action according to the present environmental state, and continuouslyperforming training and learning through an evaluation made by Critic, thereby obtaining an optimal control strategy to ensure that the unmanned vehicle can be kept at a certain safe distance away from the front and back vehicles and implement automatic low-speed running of the vehicle following the front vehicle under an urban congestion working condition. According to the deep reinforcement learning-based low-speed vehicle following decision-making method, the driving comfort is improved, the traffic safety is also ensured, and the clarity of a congested lane is further improved.

Description

technical field [0001] The invention relates to the field of automobile automatic driving, in particular to a vehicle low-speed car-following decision-making method based on deep reinforcement learning. Background technique [0002] With the development of cities and traffic, traffic congestion often occurs in the main road sections of morning and evening peaks in many cities. People's driving behavior is mainly in a stop-and-go state when vehicles are congested. Driving on congested roads for a long time will cause drivers to feel irritable. And driving fatigue, resulting in negligent or excessive driving behavior, resulting in collisions, rear-end collisions and other traffic accidents, further aggravating urban road traffic congestion, and bringing great inconvenience to people driving. [0003] The existing car-following technology based on advanced driving assistance technology mainly builds a car-following decision model based on the distance between the front and rear...

AI Technical Summary

Problems solved by technology

The movement of cells is discrete in space and time. This method is mainly used in traffic simulation, which has a large gap with driving in the actual environment.

Images